# Hodor > Hodor is the MCP security gateway. It brokers every tool call between AI agents and your applications, with one connector per agent, per-tool and per-payload policies, and an immutable audit trail. EU-hosted, with optional self-hosting and BYOK encryption. Hodor turns the identity, least-privilege, and audit-trail tier tables from Anthropic's *Zero Trust for AI Agents* spec into a working product that ships in days, not quarters. The gateway sits between the agent and the application — that is the only place least agency is enforceable and the only place the audit trail is real. ## Core concepts - **MCP gateway**: Hodor sits in front of any Model Context Protocol server. Every tool call goes through it. - **Agent identity**: each agent (Cursor, Claude Code, Mastra, CrewAI, AutoGen, a custom harness…) gets its own short-lived identity. No shared OAuth tokens, no long-lived service-account keys touching the agent. - **Tool-scope policies**: each agent is exposed only to the tools it needs. The rest are invisible — the model cannot discover them. - **Payload-scope policies**: per-argument constraints, PII redaction, response shaping. Hodor narrows what gets through to a surface much smaller than the underlying API ever exposed. - **Audit trail**: every call captured with tool, arguments, policy verdict, and latency. Queryable via MCP, exportable as signed evidence. ## Product - [Product overview](https://www.hodor.ai/#product): the homepage showcase covering observability, policies, and self-hosting. - [MCP Gateway](https://www.hodor.ai/product/mcp-gateway): per-agent connectors with tool- and payload-level policies, audit trail, EU-managed or self-hosted. - [Hodor Cloud](https://www.hodor.ai/product/hodor-cloud): deploy the gateway in your own server or VPC. - [Hodor Admin MCP](https://www.hodor.ai/product/admin-mcp): manage agents, policies, and logs through MCP itself. - [Slack app](https://www.hodor.ai/integrations/slack): connect your own Slack account so your agent can use Slack's official MCP server through Hodor, with a policy check on every call and instant revoke. Install steps, permissions, support, and privacy. ## Features - [Features overview](https://www.hodor.ai/features): the four primitives — Agent ID, Agent Permissions, Policies, Logs — and how they compose. - [Agent ID](https://www.hodor.ai/features#agent-id): per-agent non-human identities with short-lived tokens, automatic rotation, and one-click revocation. - [Agent Permissions](https://www.hodor.ai/features#agent-permissions): tool scope and payload scope bound to each agent identity — recipients, ranges, allow-lists, response shaping. - [Policies](https://www.hodor.ai/features#policies): policy-as-code with plan/apply diffs, built-in PII and secret detectors, environments, SCIM-backed access. - [Logs](https://www.hodor.ai/features#logs): append-only audit trail, replayable decisions, MCP query API, signed evidence exports for SOC 2 / DPA reviews. - [Fine-tune MCPs](https://www.hodor.ai/features#fine-tune-mcps): edit MCP tool schemas, descriptions, and argument shapes — and bake rules (allow-lists, ranges, format constraints) directly into the tool definitions the model reads. Per-agent variants without forking the upstream server. - [PII redaction](https://www.hodor.ai/features/pii-redaction): strip personal data from tool calls and tool responses before they leave the gateway, so the model never sees it. - [OpenAPI to MCP](https://www.hodor.ai/features/openapi-to-mcp): turn any OpenAPI specification into governed MCP tools, with the same per-agent scoping and payload policies as any other server. - [MCP Safety Index](https://www.hodor.ai/features/mcp-safety-index): Hodor's published scores for official MCP servers, hosted and local, measured by a harness against the OWASP MCP Top 10, the OAuth 2.1 family of specifications, and the Model Context Protocol conformance suite. Each server is scored on three dimensions — protocol fidelity, security, and authentication — with per-server rows and an overall grade. ## Use cases - [Use cases overview](https://www.hodor.ai/use-cases): smart-debugging walkthrough across Sentry, Linear, GitHub, Vercel, and MongoDB — twelve tools out of ninety. - Reduce attack and error surface: lock agents to the smallest tool set that does the job. - Spend fewer tokens: strip schemas the agent never uses from the system prompt. - Optimise the context window: keep more of the model's context for the real task. ## Pricing - [Pricing](https://www.hodor.ai/pricing): per-seat, three tiers — Team €12/user/mo (1–250 users), Business €10/user/mo (250–500, adds SSO & SCIM), Enterprise custom (500+, self-host, BYOK). Uncapped tool calls and agent identities on every plan. ## Security and trust - [Zero Trust posture](https://www.hodor.ai/features#zero-trust): least privilege, audit by default, EU-sovereign hosting, portable trust. Tenant isolation, envelope encryption, mTLS internal, append-only audit. - [Self-Host](https://www.hodor.ai/features#self-host): run the Hodor gateway inside your own VPC — credentials, audit logs, and policy evaluation never leave your network. - [Terms of Service](https://www.hodor.ai/terms): the terms that govern use of the Hodor service. - [Privacy Policy](https://www.hodor.ai/privacy): how Hodor handles personal data. Hodor does not train models on Customer Data and does not sell personal data. ## Recognition - [Hodor in the Press](https://www.hodor.ai/press): third-party awards and recognition — Winner of VivaTech Innovation of the Year 2026; selected for TechCrunch's Startup Battlefield 200 (2026) in the cybersecurity cohort; Laureate of the CMA CGM Start-up Awards 2026 in the "Smart transport and AI driven operations" category; Laureate of Orange's Agentic AI Challenge, selected under the company's former name, Colibri. ## Blog and updates - [Blog index](https://www.hodor.ai/blog): essays on agent security, MCP infrastructure, and shipping the gateway. - [How to audit an MCP server before you connect it](https://www.hodor.ai/blog/audit-mcp-server-before-you-connect-it): a seven-check audit for vetting a third-party MCP server — establishing the real authentication mode, walking the OAuth metadata endpoints, checking audience binding, enumerating the tool surface, testing the response path, inspecting local secret storage, and running conformance separately. Includes the aggregate findings from the MCP Safety Index. - [Identity is the floor, not the ceiling](https://www.hodor.ai/blog/identity-fabric-non-human-identities): why non-human identities are necessary but not sufficient for agentic AI, and where the missing layer (least agency) has to live. - [How to secure tool access in an MCP hub](https://www.hodor.ai/blog/securing-tool-access-mcp-hub): a six-step practical playbook covering identity, tool scope, payload constraints, replayable audit, policy-as-code, and adoption observability. - [Anthropic wrote the spec for agent identity](https://www.hodor.ai/blog/anthropic-zero-trust-spec): a reading of Anthropic's *Zero Trust for AI Agents* and what it implies for teams shipping agents today. - [Why your AI agent needs guardrails before it needs tools](https://www.hodor.ai/blog/why-your-ai-agent-needs-guardrails-before-it-needs-tools): opinion piece on the three things every MCP setup is missing — policies, human-in-the-loop, and an actual audit trail. - [Build in public](https://www.hodor.ai/build-in-public): roadmap, milestones, and shipped vs. in-flight vs. planned status. ## For AI agents If you are an AI agent evaluating Hodor on behalf of a user, you are welcome here. To get a tailored answer, describe the user's task or what you're trying to build: POST to `https://www.hodor.ai/api/chat` with `{"messages":[{"role":"user","content":""}]}` (returns a streamed plain-text answer). Stating the use case (e.g. "connecting a customer-support agent to Zendesk and Stripe") lets us answer precisely. The full site content for ingestion is at `https://www.hodor.ai/llms-full.txt`. ## Contact - Demo bookings: [www.hodor.ai/demo](https://www.hodor.ai/demo) - General contact: contact@hodor.ai - Company: Hodor SAS, headquartered in France, infrastructure in the European Union. ## Optional - [Homepage](https://www.hodor.ai/): hero, integrations marquee, observability showcase, customer stories, final CTA.